Note: "permalinks" may not be as permanent as we would like,
direct links of old sources may well be a few messages off.
On Fri, Oct 19, 2012 at 11:04:29AM -0500, Andrew Martin wrote: > Hello, > > I've recently upgraded my Pacemaker + Corosync cluster to Pacemaker 1.1.8 and Corosync 2.1.0. One of the resources I have is a DRBD resource (DRBD 8.3.11). After performing the upgrade, I noticed that crm-fence-peer.sh was returning the following error state (defined here: http://www.drbd.org/users-guide/s-fence-peer.html) when I would disrupt the DRBD replication link, even though the location constraint was being successfully added to the CIB: > Connection to the peer node failed, peer could not be reached. (exit code 5) > > I discovered that my CIB XML does not contain the "ha" attribute for the <node_state> tag, which crm-fence-peer.sh uses to determine if the peer node is reachable: > if ! echo "$state_lines" | grep -v -F uname=\"$DRBD_PEER\" | grep -q 'ha="active"'; then > > This ha attribute appears to still be used in the latest version of DRBD as well ( http://git.linbit.com/gitweb.cgi?p=drbd-8.3.git;a=blob;f=scripts/crm-fence-peer.sh;hb=HEAD ). > > I updated the script to check the status of the crmd parameter instead (repeat this same modification in all of the places where ha="active" is used): > if ! echo "$state_lines" | grep -v -F uname=\"$DRBD_PEER\" | grep -q 'crmd="online"'; then > > After making this change, crm-fence-peer.sh returns the expected exit code (4) > Peer’s disk state was successfully set to Outdated (or was Outdated to begin with). Sounds good. Thanks for reporting. > Does crm-fence-peer.sh also run "drbdadm outdate res" on the peer node (it doesn't seem to be working for me), No. > or is that unnecessary in this scenario because the CIB constraint will prevent the DRBD resource from starting elsewhere? Yes, that's the idea. -- : Lars Ellenberg : LINBIT | Your Way to High Availability : DRBD/HA support and consulting http://www.linbit.com